The MCP Playbook: How I Wired Cursor to My Entire Production Stack

Your Cursor IDE knows about your current file. Maybe your project. But does it know about your running Docker containers, your database schemas, your deployment pipeline, and your seven microservices talking to each other over an internal Docker network?

Mine does. I built feldhege-stack-mcp — a custom Model Context Protocol server that connects Cursor to my entire production infrastructure: a Next.js frontend, seven FastAPI/Node.js microservices, three dedicated PostgreSQL databases, a Qdrant vector store, and an Ollama instance, all running in Docker on a VPS.

In this talk, I'll show you the MCP server architecture, walk through the key tools and resources I exposed, and demonstrate how this changes the development experience. When Cursor understands your infrastructure, it stops generating code in isolation and starts making architecture-aware suggestions. It can query your vector database schema before building a RAG pipeline. It knows which service runs on which port. It understands your nginx routing.

You'll leave with a mental model for designing your own production MCP server — and a clear list of what not to expose. I'll also show what went wrong: the context window limits that forced me to be selective about what to expose, the latency issues with remote MCP connections, and the security decisions I had to make.

Minimal slides — primarily live demo and terminal sessions.